A Bayesian Treed Model of Online Purchasing Behavior Using In ...

Informations

• Publié par: axycozor
• Nombre de lectures: 93
• Langue: Français

Extrait

           

   

   

 

 

A Bayesian Treed Model of Online Purchasing Behavior

Using In-Store Navigational Clickstream

Wendy W. Moe Assistant Professor of Marketing University of Texas at Austin 2100 Speedway, CBA 7.216 Austin, TX 78703 [e-mail]  [phone]   Hugh Chipman Associate Professor of Statistics University of Waterloo   Edward I. George Professor of Statistics University of Pennsylvania, Wharton   Robert E. McCulloch Professor of Econometrics and Statistics University of Chicago           April 2002

A Bayesian Treed Model of Online Purchasing Behavior Using In-Store Navigational

Clickstream

  ABSTRACT   Internet clickstream data allows online retailers to observe their customers as they click from page to page. The ability to infer a customers motivation and search strategy (and therefore his/her likely response to in-store experiences) from observed clickstream data has tremendous benefits for online marketers when segmenting and targeting these individuals. Toward that effort, we develop a Bayesian treed model that simultaneously (1) groups shoppers based on patterns observed in their online navigational clickstream and (2) examines their purchasing decision as a function of in-store experiences. Bayesian treed models are ideal for this research problem as they can identify structure across observations by accommodating two different sources of variance, using variables for segmentation as well as for prediction within segments. First, we account for heterogeneity by branching the tree into segments depending on how each observation scores along navigational measures. Then, variance in purchasing behavior within each terminal node of the tree is modeled to examine the purchasing processes driving each visit type. Alternative methods such as a latent class logit model are also considered but are seen to be significantly less effective in representing the patterns found in the data.

   

1  

A Bayesian Treed Model of Online Purchasing Behavior Using In-Store Navigational Clickstream  

Introduction

Despite the rapid growth in e-commerce, online purchasing conversion rates have remained

chronically low. For a typical online retailer, only 1-2% of visits convert into purchases

(Brownlow 2001). This implies that over 95% of visits to a given retail site are made by

shoppers who are merely browsing. Increasing these conversion rates has been a primary focus

of many e-commerce marketers. But before we can begin exploring methods to achieve this

goal, we must first gain a clearer understanding of the differences between browsing and buying

visits in terms of their underlying motivations and purchasing processes. Additionally, we must

also develop methods and measures with which marketers can differentiate between these visits.

 

For example, imagine a student who needs to buy a very specific textbook for class. This

individuals behavior at an online bookstore will be very different from that of someone who

may merely be browsing for the entertainment value of window shopping, both in terms of

navigational pattern as well as purchasing response to various aspects of the store visit. In a

previous study, online shoppers were found to vary dramatically in terms of their motivations

and search strategies and could be identified by their unique clickstream patterns (Moe 2002).

In this paper, we develop methods and measures not only to differentiate between different types

of store visits but also to model the differences that exist between purchasing processes.

 

Internet clickstream data allows online retailers to observe their customers as they click from

page to page. The ability to infer a customers motivation and search strategy (and therefore

   

2  

his/her likely response to in-store experiences) from observed clickstream data has tremendous

benefits for online marketers when segmenting and targeting these individuals. Toward that

effort, we develop a Bayesian treed model (Chipman, George, and McCulloch 2002, hereafter

CGM) that simultaneously (1) groups shoppers based on patterns observed in their online

navigational clickstream and (2) examines their purchasing decision as a function of in-store

experiences. Bayesian treed models are ideal for this research problem as they can identify

structure across observations by accommodating two different sources of variance, using

variables for segmentation as well as for prediction within segments. First, we account for

heterogeneity by branching the tree into segments depending on how each observation scores

along navigational measures. Then, variance in purchasing behavior within each terminal node

of the tree is modeled to examine the purchasing processes driving each visit type. Alternative

methods such as a latent class logit model are also considered but are seen to be significantly less

effective in representing the patterns found in the data.

  

Previous studies (Bucklin and Sismeiro 2000, Novak, Hoffman, and Yung 2000) have explored

navigational behavior in terms of the effects of page-depth (the number of pages viewed in a

shopping session) and session-duration (the amount of time spent during an online session).

However, these studies and the measures used in them have ignored thecontentof the pages

viewed and the effect of this content on purchasing. But without knowingwhatthe shopper sees

from page-to-page, it is difficult to understand the effects of these pages in terms of their role in

the navigational process and their influence on the purchasing decision. In this paper, we

directly examine the role of page content by using a number of within-session metrics that are

suitable for most online retailers and efficiently characterize the unique patterns found in

   

3  

navigational clickstream data. Used in our Bayesian treed model, these measures will allow us

to differentiate between store visits and hence model purchasing accordingly.

 

Additionally, it has been shown that consumers often construct their preferences during their

online shopping session according to the page content encountered during the session (Mandel

and Johnson 2000). But despite this apparent relationship between in-store navigation and

preference construction, very little research, if any, has been conducted to study the effect of

pageviews on whether or not the individual purchases. Therefore, at the root of our Bayesian

treed model is a logit model of purchasing that examines the effect various aspects of a shoppers

in-store experience (characterized by the pages and content viewed) have on the buying decision.

But since the model also allows for heterogeneity in motivation and search strategy across store

visits, we can measure differences in these pageview effects on purchasing across different types

of shopping sessions.

 

In the next sections, we will review some of the relevant literature on search behavior and

discuss the implications they have for modeling in-store navigational behavior and purchasing.

Then we will provide a thorough discussion of the type of data and measures that we will use in

this study. From there, we describe Bayesian treed models and apply the method to our online

shopping problem.

 

Online Search Behavior

Before we begin developing a model of navigational patterns and purchasing probabilities, we

first discuss some of the ways in which online search behavior is expected to vary. Specifically,

   

4  

we differentiate in-store behavior along two dimensions: (1) goal-directed versus exploratory

search and (2) stage of purchasing process.

 

Previous offline research has dichotomized search behavior into goal-directed versus exploratory

search (Janiszewski 1998). Others have also applied such a dichotomy to online behavior when

segmenting shoppers (Moe 2002) or conceptualizing an internet users state offlowffam n( oH

and Novak 1996). In these studies, goal-directed search refers to behavior for which the shopper

has a specific or planned purchase in mind. The objective of search, in this situation, is to gather

relevant information regarding a specific purchasing decision (Brucks 1985, Wilkie and Dickson

1985). Therefore, it would be expected of a goal-directed shopper that in-store navigation is

very focused around a given purchasing decision.

 

Assumption 1: Directed search behavior can be characterized by a high degree offocus  exhibited in a shoppers store visit. The more focused the store visit is around a specific product purchasing decision, the more likely the shopper is engaging in goal-directed search behavior.

Exploratory search, on the other hand, refers to behavior in which the consumer is less

deliberate, less focused, and perhaps not even considering a purchase. Instead, navigation tends

to be undirected and stimulus-driven rather than goal-driven (Janiszewski 1998). In this case, the

shopper derives hedonic utility from the shopping process itself and the environmental stimuli

encountered (Babin, Darden, and Griffin 1994, Sherry, McGrath, and Levy 1993, Hirschman

1984). This type of behavior is also sometimes referred to as browsing or ongoing search. Since

ongoing search is not motivated by any specific decision making need, in-store activities are

expected to exhibit less focus around a specific product decision and more browsing variety

across different products and product categories.

   

5  

 

Assumption 2: Exploratory search behavior can be characterized by a low degree of focusin a shoppers  more a shoppers Thestore visit around a specific product decision. navigational patterns reflect a highly varied set of product considerations, the less likely the shopper has a planned purchase in mind and the more likely the individual is engaging in exploratory search behavior.

One key differentiating factor between goal-directed and exploratory search behavior is the

shoppers purchasing orientation at the start of the store visit. For example, exploratory shoppers

do not necessarily enter the store with an intended purchase in mind. In contrast, goal-directed

shoppers often have a very specific product purchase in mind and, as a result, tend to be

predisposed to purchasing upon entering the store. Because of these differing motivations, we

expect goal-directed shoppers to be more likely to purchase than exploratory shoppers.

 

Proposition 1: Goal-directed shoppers, as identified by the level of focus in their navigational patterns, are more likely to purchase than exploratory shoppers.

In addition to differences in navigational behavior and in the overall propensity to purchase

between goal-directed and exploratory search, there are also significant difference in the

purchasingprocess Thethat each entails.more commonly studied search behavior in marketing 

research tends to be goal-directed search where the individual goes through a process of forming

a consideration set, deliberating the options, and making a final purchasing decision based on the

attributes of each product in the set (Kardes et al 1993, Shocker et al 1991, Ratneshwar et al

1991, Barsalou 1991). The top panel of Figure 1 illustrates the purchasing process of a goal-

directed shopper.

 

 

   

FIGURE 1 HERE (The Purchasing Process)

6  

Navigational patterns for a goal-directed shopper will depend on the stage of the purchasing

process. Initially, in-store navigation will aim to construct a consideration set by searching

across products within a particular category. Once this is done, the behavior turns to the

deliberation of options where each item in the consideration set is carefully evaluated based on

their product attributes. Depending on the outcome of the deliberation process, purchase may or

may not occur.

 

Putsis and Srinivasan (1994) studied the pre-purchase deliberation process described here and

proposed that customers endeavor to gather and accumulate information during the pre-purchase

deliberation process. When enough relevant information is acquired to surpass a threshold,

purchase occurs (Putsis and Srinivasan 1994, Moe and Fader 2001). In many cases, goal-

directed shoppers have already made the decision to buy in a particular product category, and it

is simply a matter of choosing a specific product in that category. This final decision hinges on

the shoppers ability to gather enough product information pertaining to that purchase. Early

stage shoppers (i.e., those who are still building their consideration set) have yet to acquire

enough information to make such a decision and are therefore less likely buy. Later in the

process, after sufficient product-related information has been collected, more in-depth

deliberations are possible, making purchasing more likely.

 

Proposition 2: Goal-directed shoppers in the later stages of deliberation are more likely to purchase than those goal-directed shoppers who are in earlier stages of the purchasing process.

Goal-directed shoppers are influenced by different factors depending on the stage of the decision

process. Early stage goal-directed shoppers are focused on identifying eligible products and

constructing their consideration set. Therefore, these shoppers will respond positively to any in-

   

7  

store experiences that aid in this process. Once a shopper enters the deliberation stage, in-store

activities that allow the shopper to more carefully examine a product will increase the

individuals likelihood of buying. This could include reviewing detailed product information

previously seen during the consideration set formation process but not necessarily closely

examined.

 

 

 

Proposition 3a: Early stage goal-directed shoppers will be positively influenced to purchase by experiences that help them identify eligible products and construct their consideration set.

Proposition 3b: Late stage goal-directed shoppers are positively affected by reviewing any product information that allows them to more closely examine a product in the consideration set. This information has likely been viewed earlier when the consideration set was being constructed.

Exploratory shoppers go through a very different purchasing process (Figure 1). Exploratory

search starts not with the objective of forming a consideration set but with the voluntary

exposure to environmental stimuli (e.g., viewing a wide variety of products and product

categories). One of two outcomes could result from this activity. First, the stimulus encountered

may not be of any interest to the shopper and therefore would not influence the shopper to buy.

In this case, the shopper exits the store without purchasing. However, the shopper may also

encounter interesting information that may stimulate the shopper to consider a purchase. This

would drive the shopper to further examine and evaluate a specific product, potentially leading to

what is defined as an unplanned or impulse purchase (Janiszewski 1998, Jarboe and McDaniel

1987). For example, imagine an exploratory shopper in an online bookstore. As the shopper

browses across categories and across books, he/she will encounter many new products and much

product information. If these products are uninteresting to the shopper (non-positive influence),

then the shopper will ultimately leave the store without buying. If, however, the shopper comes

   

8  

across a book that looks interesting, he/she may examine the specific book more closely by

acquiring detailed product information. Shoppers who reach this later stage of exploratory

search are more likely to buy than early stage exploratory shoppers who have yet to find a

product worth considering.

 

Proposition 4: Exploratory shoppers in the later stages of deliberation (i.e., those who are examining and evaluating product details) are more likely to buy than those who are still just browsing.

We now turn to the factors that influence purchasing among exploratory shoppers. In general,

the probability that these stimulus-driven shoppers will find an item worth considering is a

stochastic process, where each product-related experience has the potential of generating interest

and perhaps resulting in a purchase. We therefore expect that increased exposure to products

available at the store increases the probability that the shopper will buy. As an illustrative

example, imagine a child in a toy store. The more new toys the child is exposed to in the toy

store, the more likely one of the toys will be irresistible and ultimately purchased.

 

Proposition 5: The more exposure an exploratory shopper has to products in the store, the more likely he/she will be to buy. That is, purchasing is positively influenced by increased product-related experiences.

After a product has been identified as a possible purchase, the shopper enters the deliberation

stage. In this stage, the shoppers objective is to further examine a product previously viewed

and deemed interesting. Exploratory shoppers in this stage are similar to early stage goal-

directed shoppers in that repeat examination of product information helps the shopper further

evaluate the purchasing decision, increasing the likelihood that he/she will reach a final decision

and buy.

   

9  

 

Data

Proposition 6: Late stage exploratory shoppers (who are deliberating a purchase) are positively influenced to buy when reviewing product information for items previously identified as potential purchases.

The Store Site 

The context of our study is an online retailer that sells nutrition products such as vitamins,

weight loss aids, body-building supplements, etc. The range of their product offerings provides a

mix of customer types ranging from casually health conscious consumers, interested in buying

daily vitamins and nutrition supplements, to health and body-building fanatics, looking for

performance enhancers and protein supplements. These shoppers vary dramatically in terms of

their objectives, involvement levels, and expertise, which should lead to very different shopping

strategies as reflected by their page-to-page behavior at the site. A relatively small and new site,

the store experiences roughly 5,000 to 10,000 visits a month, approximately 80% of which are

made by unique visitors. Their conversion rates, the proportion of visits that end with a

purchase, are in line with the industry averaging slightly less than 2%.

 

From the sites homepage, the shopper has a number of options. First, the store visitor may

choose to login and manage their account. This includes activities such as registering as a new

user, updating your personal profile, monitoring the status of a purchase, etc. Second, the visitor

may view informational pages. A common practice of e-commerce retailers is to provide

community areas on their sites or advice columns to help shoppers learn more about nutritional

and health issues related to the products that they sell. Visitors may also access the customer

service pages for information about the shipping, return policies, the company, and privacy

issues before deciding whether or not to transact with the site. Last, but definitely not least, are

   

10